Storage Proxy: Maintain metrics in response handler classes

[eliransin]

There is a known problem introduced by 971711a, the problem is that when some callbacks for a created response handler are called from another scheduling groups an imbalance is introduced to statistics which are now instantiated per scheduling group. This can be easily solved by "remembering" the stats of the scheduling group where the response handler was created. This will mask out another problem where certain operations are done outside of the "native" scheduling group. This is a complex issue that needs to be addressed in separate commit (see 971711a commit message).

Fixes #5797

[eliransin]

An older version of the patch have already been reviewed by @gleb-cloudius and @psarna , the patch title in the mailing list is: [PATCH] Storage Proxy: Maintain metrics in response handler classes The main change since the older patch is the introduction of paxos_executor and paxos stats to the structure.

[slivne]

@eliransin / @psarna / @gleb-cloudius is this till relevant if so lets review / merge

[eliransin]

We will need this fix. @gleb-cloudius can you review?

[eliransin]

This was done already